M360 Eurasia came to Samarkand for the first time this year — a city that spent centuries at the center of Silk Road trade and is now building a different kind of hub: one where digital sovereignty, AI ambition, and the infrastructure those ambitions depend on are all being assembled at the same time.
M360 Eurasia brought together operators, regulators, and technology players from across Central Asia, the CIS region, and beyond. Two days of conversations moved between the aspirational and the deeply practical, and below I’ll reflect on some of the key themes.
Opensignal published its own analysis of the region's mobile trajectory ahead of the event — Central Asia: Mobile Future from 5G Hype to Measurable Network Excellence — which informed several of the sessions I participated in. Here is what stood out from the conversations on the ground.
Tech sovereignty is not a slogan anymore — it is a procurement decision
The framing from GSMA Chief Regulatory Officer John Giusti set the tone early: 5G progress needs policy. But what struck me more was how concretely that is playing out in this region.
Kazakhstan and Uzbekistan are both building sovereign AI infrastructure — local language models, governed cloud, and in Uzbekistan's case, a new independent regulator. VEON, which operates under the Beeline brand across the region, is integrating AI built on local language stacks. The rationale is clear: general-purpose models get domain-specific data wrong. The Open Telco AI initiative is working on exactly this — telco domain benchmarks and fit-for-purpose AI rather than off-the-shelf.
The numbers behind Uzbekistan's transformation are worth pausing on. Between 2017 and 2026, mobile broadband coverage went from 35% to 99.5%, fixed broadband from 30% to 98.2%, and international bandwidth capacity from 94 Gbps to 5,000 Gbps. Internet penetration is now at 94.2%. That is the base on which sovereignty ambitions are being built.
Uzbekistan's Minister of Digital Technologies, HE Sherzod Shermatov, outlined the next layer: tax exemptions for IT companies, an independent Telecommunications Regulatory Agency established in 2025 aligned with WTO and ITU standards, and a dedicated AI education program. The country is also actively pitching Karakalpakstan as a hyperscale data center location — 0% corporate, property, and social tax until 2040, government-guaranteed external infrastructure including power and connectivity, 9,300 MW of wind capacity, and a targeted 2.3 EFLOPS of AI compute powered by NVIDIA, backed by Presidential Decree PD №189.
But as Giusti put it — and it is worth stating plainly — balance matters here. Supporting local capabilities cannot mean fragmenting the ecosystem. Building resilience cannot come at the cost of interoperability. And policy frameworks that shift unpredictably do not attract the long-term capital this region needs.
Forty-plus countries globally are now building some form of sovereign AI infrastructure. The risk is not ambition — it is fragmentation that raises costs for operators and slows down the innovation cycle.
The traditional telco business model is under real pressure
VEON CEO Kaan Tezoglu made a case I have heard versions of before, but rarely stated this directly: stop selling raw data, start delivering meaningful services.
VEON operates across five countries, has 150 million SIM cards, and frames its opportunity around 1,440 minutes a day — every minute a potential touchpoint for financial services, health, entertainment, and more. They call it the "digital operator" model. It is not new as a concept, but what is changing is the urgency. Revenue from traditional connectivity is not going to sustain the investment cycle.
Beeline Kazakhstan made a similar point from a different angle, and as the CEO Jabbor Kayumov put it: "Traditional telco is not something we can afford." He cited that the operator built 26 digital products in-house and now generates 20% of revenue from them. Digital is not a side project — it is the survival strategy.
The pressure is not unique to this region. As Opensignal has noted previously, 5G monetization remains challenging globally — the industry is still searching for a definitive consumer use case beyond faster speeds. Operators that stay in the lane of raw connectivity will face continued margin compression. The ones that move toward meaningful services — and build the data assets and trust infrastructure to do it — have a different story to tell.
Satellite aiming to fill the coverage gaps
Beeline Kazakhstan’s partnership with Starlink is a practical response to a geographic reality: terrestrial networks cover only 20% of Kazakhstan's territory. It is the ninth largest country in the world by land area. Waiting for ground-based infrastructure to close that gap is not a viable option.
Lourdes Montenegro from AST SpaceMobile laid out a more specific case. AST has 50-plus MNO partners representing approximately 3 billion subscribers, rights to 80-plus MHz of spectrum in the US, and satellites capable of tuning across 1,150 MHz of low- and mid-band spectrum globally. Their Direct to Device (D2D) model operates on two tracks:
- using leased IMT terrestrial spectrum from MNO partners under ITU Radio Regulation 4.4,
- or licensed Mobile Satellite Service spectrum in L or S band.
The device availability and cost barriers that stalled earlier iterations of this idea are coming down. But it is important to note that the D2D model comes with its own complexities — spectrum coordination, resource allocation, moving beyond SMS to live data channels — and these have to be solved before we can assess real world implications.
AI: the gap between strategy and infrastructure
Day 2 shifted to AI — and the gap between national AI strategies and the infrastructure needed to support them was visible throughout.
Huawei's keynote put a useful frame around where the industry is headed. The progression they mapped: PC internet (1 billion users, 1 hour per day) → mobile internet (10 billion users, 6 hours per day) → agentic internet (1 billion-plus agents, 24 hours per day, always on). The implication for operators is that connectivity is no longer about connecting people to services — it is about being the infrastructure that agents run on. The network that cannot deliver deterministic, always-available performance is not part of that picture.
NTT DOCOMO offered one of the more detailed pictures of what an operator-led AI strategy looks like at scale: 110 million members on their data platform, an EBITDA target of $26 billion by 2030, and AI applied across everything from network optimization to behavioral prediction and out-of-home advertising. Their 6G vision is explicitly AI-centric. They are also building an AI-ready data center in central Tokyo, scheduled for completion in 2029, and are the third-largest data center provider globally.
Yandex is taking a different approach — rebuilding around the idea that the internet gateway itself is shifting from browser to AI assistant. They are piloting in Turkey, with localized AI for everyday users, not just engineers. The monetization model: ARPU uplift through premium AI bundles, lower churn through daily engagement.
But the most grounding perspective came from Aamir Ibrahim, CEO of Jazz (Pakistan). Pakistan has 250 million people, 90 million of whom still do not have internet access. His point was simple: you cannot have a meaningful conversation about AI without first connecting people. And when you do connect them, relevance matters — affordable access, local language, practical use cases that fit real lives. JazzCash has 60 million customers, built on the insight that traditional credit histories excluded most of the population. Telco data can fill that gap — but only if the data is governed and kept within the country.
The AI sovereignty conversation in the room followed a related logic: building local capabilities without cutting off from the global ecosystem. Turkey's telecom regulator BTK framed it well: sovereignty is not about isolation — it is about trusted capabilities, resilient infrastructure, and the ability to deploy local language models while remaining part of the global ecosystem. Resilience and trust are becoming as important as sovereignty itself.
The infrastructure fundamentals cannot be skipped
In the session I moderated — Always-On Central Asia: 5G, Data Centres, and Infrastructure Reliability — I used Opensignal data to frame the central tension: there is a persistent gap between what operators invest in networks and what users actually experience. That gap is the policy problem this region needs to solve.
Daniel Pataki, Chief Regulatory Officer and EU Affairs at Cellnex Telecom, made the case for the neutral host model as a structural answer to densification — operators sharing passive infrastructure rather than duplicating Capex. The preconditions for that model to work are regulatory: tower access rules, co-location frameworks, spectrum policy that does not foreclose sharing. In Central Asia, most of those preconditions are still being assembled.
The Iberian Peninsula blackout in April 2025 gave the resilience discussion a concrete anchor. When the grid failed, it exposed that backup power requirements in most markets are not sized for extended outages and that energy and telecoms dependencies are underregulated. The EU has since moved on resilience obligations. For Central Asia — where power constraints are a real operational variable in markets like Tajikistan — the sequencing problem is harder: you cannot mandate resilience on infrastructure that does not have reliable power to begin with.
Perfectum in Uzbekistan is a useful marker of how fast this region can move when conditions are right. Starting from zero, they signed with Nokia in April 2024, launched 5G SA FWA in Tashkent in April 2025, and are targeting nationwide coverage by 2027 — no legacy network to work around. As their CEO put it: "no legacy, no cry." Straight to 5G standalone, no technical debt to manage around.
What I am taking away
- The cost of connectivity in this region is still a barrier. Not just for end users but for startups trying to build AI applications. When edge compute costs $10,000 a month, the innovation ecosystem does not develop equally.
- The "digital operator" pivot is real, but it requires data trust infrastructure that most operators have not built yet. The operators that figure out governance before regulation forces it on them will have an advantage.
- The satellite conversation has moved to being part of operators’ strategy. The device compatibility question is largely resolved. What remains is the commercial model, spectrum policy, and how terrestrial and non-terrestrial networks get managed as a single system.
Infrastructure resilience is key. The Iberian blackout should be a forcing function for Central Asian regulators, not a distant European lesson. The preconditions for resilience — backup power standards, neutral host frameworks, cross-border spectrum coordination — need to be written into license conditions now, before an event tests networks that are still being built.
One exchange stayed with me more than most. In a sideline interview at the event, Minister of Digital Technologies Sherzod Shermatov was asked about the risk of AI-caused unemployment. His answer reframed the question entirely: "It's no longer AI vs humans — but humans who can use AI vs humans who cannot." He pointed to what Uzbekistan is actually doing about it: more than 1 million young people have already completed the initial phase of a national AI education program focused on prompt engineering and practical skills. The country partnered with Coursera to expand access to generative AI courses and now ranks first globally in active Coursera learners relative to workforce-age population. More than half of those completing Gen-AI courses are women and girls — a figure Coursera itself highlighted in a recent International Women's Day report.
Those numbers are unusual enough to deserve attention. The government's AI strategy is framed around inclusivity — women, low-income families, people with disabilities — not just productivity or GDP contribution. Whether that education investment can keep pace with how quickly jobs are changing is a genuine open question. But it is at least the right question to be asking, and asking it out loud in a national policy context is not nothing.
The startup competition told the same story from a different angle. I had the chance to judge the startup competition at Pitch Day: Uzbekistan — a joint initiative between IT Park Uzbekistan, IT Ventures, ITU, and the GSMA that gave early-stage companies the chance to pitch in front of a global investor audience. The quality of ideas on display was its own data point about where this ecosystem is heading. Most of the companies pitching were actively using AI, not theorizing about it.
That is what I took from Samarkand: a region where the ambition is large and, in places, the execution is starting to match it. The gaps are real — in coverage, in regulatory maturity, in the distance between network investment and user experience that Opensignal measures. But the conversations were honest, and grounded in reality and that is always a good start.
The editorial views expressed in this article are solely those of the author and do not necessarily reflect the views of Opensignal.
Opensignal Limited retains ownership of this insight including all intellectual property rights, data, content, graphs & analysis. Reports and insights produced by Opensignal Limited may not be quoted, reproduced, distributed, published for any commercial purpose (including use in advertisements or other promotional content) without prior written consent. Journalists are encouraged to quote information included in Opensignal reports and insights provided they include clear source attribution. For more information, contact [email protected].
